Is it possible to do "Multi-round Chatting" with InstructBLIP?

[dongmean]

Hi, thanks for the great work!

I'm wondering if InstructBLIP allows multi-round chatting, e.g., <Img, Q1, A1, Q2, ?>, so that it can support in-context learning in a dialogue. If so, which python script can I refer to? It seems the demo (https://replicate.com/gfodor/instructblip) does not allow the multi-round consecutive questions.

Thanks again for your amazing contributions.

[MiaQiu]

This may be what you are looking for: https://colab.research.google.com/github/salesforce/LAVIS/blob/main/examples/blip2_instructed_generation.ipynb

[Fym68]

Have you found a method? I also want to do a In-context learning experiment.